Wednesday, February 16, 2011

Chapter 7: Empirical Estimation of Survey Measurement Error


 

First Examination of Observational Errors vs Errors of Non Observation


-          Up to now – errors of non observation – failure of survey to measure some portion of population, and on costs of reducing those errors

-          Now – errors in data obtained – measurement errors

-          Sources

o   Interviewer

o   Respondent

o   Survey questionnaire

o   Mode of communication

-          Issues

o   Different designs are used to estimate variable measurement errors and biases

o   Researchers who conceptualize an error as a bias often seek to identify cause of error and eliminate it – those who conceptualize error as source of increased variation in survey statistics often attempt to measure it and adjust their conclusions to its presence

o   Most past work studies only one of the design features as source of measurement error

 

Laboratory Experiments Resembling the Survey Interview

 

Measures External to Survey


-          One way to estimate measurement error – compare result of survey measurement to another indicator of same characteristic.

o   Obtain external data on individual persons in survey – eg, police or hospital records

o   Compare external population parameter values with survey based estimates of same quantity.

 

Randomized Assignment of Measurement Procedures to Sample Persons

 

-          Eg, randomized assignment of interviewer to sub samples of respondents.

-          Split sample experiments – two different versions of same question

-          Split sample experiments concerning measurement error

-          Biases with interviewer behaviour

-          Interviewing procedures

-          Interpenetration for estimation of variable measurement error

o   Interviewer effect

§  White interviewers appear to influence more moderate racial attitudes among black respondents who would give more radical answers to same question asked by black interviewer

o   Ordering of questions or response categories

-          Practical problems with randomization as a procedure for estimating measurement error

 

Repeated Measurements of same persons


-          Same questions

-          Multiple measures of same concept (in same survey)

-          Direct questioning of respondent about their understanding of question

-          Re-interviews to estimate measurement error

-          Multiple indicators of same characteristic administered in same questionnaire

-          Measuring characteristics of the interview itself

-          Direct measures of cognitive steps in response formation

 

Summary of Individual Techniques of Measurement Error Estimation

 

-          Replication of identical measures over trials

-          Replication of indicators of same concept within trials

-          Randomization of measurement procedures to different persons

-          Collection of correlates of measurement error

 

 

Combinations of Research Features

 

 

No comments: